Parcourir les annonces Parcourir les étudiants Publier une annonce Ajouter un CV

Dynamic forest carbon maps from very high resolution satellite data

Date de début
Date de fin

University of Copenhagen, Kayrros (a Paris based EO startup), the Laboratoire des Sciences du Climat et
de l’Environnement (LSCE) and INRAE Bordeaux are looking for a PhD candidate on a joint project.
Carbon in forests is extremely important to climate but extremely difficult to measure. The problem is that
currently no suitable tool exists for a rapid assessment of forest carbon change at the fine scale. Such a tool is
crucial to inform smart management of carbon sequestration from a myriad of projects with highly diverse
forestry practices. New satellites open a window to track carbon sequestration in individual trees. The new
PlanetScope constellation of micro-satellites take a picture of every tree on Earth every day. These images
have a very high spatial resolution of 3 meters. Moreover, the daily coverage of images from PlanetScope
allows for the first time to take the pulse of carbon in forests in almost near real time
Overall aim
Apply high spatial and temporal resolution PlanetScope satellite data to produce annual forest and carbon
stock maps for a selected region in Europe
Specific aims and working steps
• Select a forest area in Europe with available plot data.
• Develop a framework for downloading/storing/mosaicing/processing multi-spectral PlanetScope
satellite data.
• Develop a framework based on deep learning (convolutional neural networks) to map forests and specific
parameters (e.g., tree density, management type, dominant species, forest age) using the capacities
of the very high spatial and temporal resolution offered by PlanetScope. Training data is derived
from national inventory plot data, own field surveys, and Lidar.
• Use above mentioned data sources to derive biomass carbon stocks at a 3-m resolution for 2018-2020.
• Programming skills, preferably in Python
• Basic understanding of satellite images and spatial analyses
• Knowledge on machine learning, preferably deep learning
Martin Brandt,
Philippe Ciais,


3 mars 2020